MPAS homepage banner
X (formerly Twitter) Instagram Vimeo Youtube Envelope

NOAA Chooses MPAS to be the Next-Generation NWS Operational Model

Authors: Eric Aligo1,2, Ligia Bernardet3, Jennifer Vogt1, Jacob Carley4, and Clark Evans3
  1. National Oceanic and Atmospheric Administration (NOAA) / Office of Oceanic and Atmospheric Research (OAR) / Weather Program Office (WPO) / Earth Prediction Innovation Center (EPIC)
  2. Science and Technology Corporation (STC)
  3. National Oceanic and Atmospheric Administration (NOAA) / Office of Oceanic and Atmospheric Research (OAR) / Global Systems Laboratory (GSL)
  4. National Oceanic and Atmospheric Administration (NOAA) / National Weather Service (NWS) / National Centers for Environmental Prediction (NCEP) / Environmental Modeling Center (EMC)

At the annual American Meteorological Society (AMS) meeting on January 26, 2026, the National Weather Service (NWS) Director, Ken Graham, announced an ambitious 10-year goal for NOAA to adopt a global, fully coupled, and convection allowing 3-km unified system using the Model for Prediction Across Scales (MPAS) for both research and operations.

Significant advancements have already been made by the NOAA Oceanic and Atmospheric Research (OAR) Laboratories with real-time global and limited area experimental model runs at varying horizontal resolutions in an atmospheric-only configuration. These runs used the standalone MPAS capability distributed and supported by National Science Foundation (NSF) National Center for Atmospheric Research (NCAR), oftentimes with additional enhancements in physical parameterizations. Experimental MPAS model runs were evaluated in various 2024 and 2025 testbed experiments covering all seasons across North America and the tropics. Figure 1 shows that MPAS performed the best on average for subjective evaluation of composite reflectivity and updraft helicity (UH) compared to flagship deterministic models at Day 1 lead times during the 2025 Hazardous Weather Testbed (HWT) Spring Forecast Experiment. Figure 2 shows that MPAS achieved significant improvements in track prediction accuracy, with notably low track errors across all forecast hours, with intensity errors highly successful beyond 24-h, for Atlantic tropical systems during the 2024 Hurricane Forecast Improvement Program (HFIP) Real-time Experiment (HREx). The Global Systems Laboratory (GSL) plans to conduct global storm-scale forecasting experiments this summer.

NOAA is moving forward with integrating the MPAS dynamical core (dycore) into the Unified Forecast System (UFS), advancing it toward a fully coupled Earth System Model (ESM) with enhanced capabilities including asynchronous input/output, improved post-processing, and a robust user support system. The Weather Program Office (WPO), Earth Prediction Innovation Center (EPIC), NWS, and OAR labs are finalizing a formal strategic plan to guide the development of a system for use in a future Rapid Refresh Forecast System (RRFS) alongside paving the way for further adoption across UFS.  

Violin plot showing distribution of subjective scores (higher is better) for Day 1 composite reflectivity and updraft helicity across five models, with median values around 5.8 to 6.3 for GSL MPAS 3km, HRRRv4, RRFS, NSSL MPAS RT, and NASA GEOS FV3.
Figure 1. Subjective skill scores for composite reflectivity and updraft helicity (UH) for deterministic models including the MPAS standalone at Day 1 lead times during the 5-week 2025 HWT Spring Forecast Experiment.
Two-panel chart showing mean track error and intensity mean absolute error for multiple hurricane forecast models across forecast hours from 12 to 120, with increasing error at longer lead times and comparison between OFCL, MPAS, HFSA, HFSB, AVNO, DSHP, and LGEM models.
Figure 2. (a) Season-mean track/position error (in nautical miles) from the Official National Hurricane Center (NHC) Forecast (OFCL) and selected models (standalone MPAS in orange), for Atlantic tropical cyclones during the 2024 HFIP real-time experiment. (b) As in (a), except for the season-mean intensity mean absolute error (in kt). In both panels, lower values indicate better forecasts. The number of forecast cycles at each lead time is depicted in bold at the bottom of each panel.